The November issue of International Taxation carried a translated paper by Jeffrey Owens and co-authors on what large language models can do in tax administration and where the risks are. Tools such as ChatGPT are most directly useful for helping taxpayers look up rules and get answers, and can also lift compliance, strengthen risk management and speed up dispute resolution. The authors caution that models give wrong or biased answers, sometimes invent content, and raise questions of transparency, privacy, security and cost. They recommend regulators define legality and accountability first, then scale up, while watching the digital divide.
News Detail
Generative AI Tools: The Potential of Large Language Models in Tax Administration
A translated research paper by Jeffrey Owens and co-authors in International Taxation assesses how large language models can be used in tax administration and where the risks lie. Models such as ChatGPT can support taxpayer queries, improve compliance, strengthen risk management and speed up dispute resolution, while their data sources and training processes shape the risks. The authors highlight inaccurate or biased answers, hallucination, opacity, data privacy, security, sustainability, the digital divide and accountability, urging policymakers to settle legality and responsibility rules before scaling up.
2026-11-02